Meta eyes cloud compute sales as AI infrastructure overbuild starts to bite

Meta is exploring selling off its surplus AI compute capacity as a cloud service, according to Bloomberg — a sign that big tech's aggressive infrastructure spending spree may be catching up with the industry. The plan is still in its earliest stages, and the company hasn't committed to any specific approach yet.

Two options are reportedly on the table. One mirrors Amazon Web Services' Bedrock model: developers pay to access AI models like Muse Spark hosted on Meta's own hardware. The other skips the model layer entirely and sells raw compute power directly. Mark Zuckerberg acknowledged the possibility at Meta's annual shareholder meeting in May, saying cloud sales were "definitely on the table" — but added that Meta hadn't acted on it yet because the company still believed it had uses for the capacity. He framed it as a fallback: "if we get to a point where we feel that we have overbuilt, then that is an option that we have." Meta's shares jumped 9.3% to $615.55 on Wednesday following the Bloomberg report, which the outlet called the biggest intraday gain since April.

The broader picture is messier than a single stock bump suggests. Meta's full-year capital expenditure projections have blown past $140 billion, and analyst Damir Tokic of Seeking Alpha warns this pivot may not actually be good news. Meta's revenue is overwhelmingly advertising-dependent — it built out AI infrastructure to serve that business, not to become a cloud provider. Tokic argues that selling compute likely signals Meta will need to slow or stop further datacenter investment, and notes the company has already borrowed around $20 billion across 2025 and 2026 while operating cash flows are declining. Meta isn't alone in this bind: SpaceX, after acquiring xAI in February, has already started renting its own excess compute to Anthropic. The risk is that if multiple AI players simultaneously realise they've overbuilt and flood the market with spare capacity, the strategy becomes self-defeating — everyone's selling to everyone else at shrinking margins.

For PC gamers and consumers, the overbuild has a direct cost: DRAM and flash chip prices have been inflated by big tech hoovering up memory and storage components to feed AI ambitions. Whether a correction in AI infrastructure spending translates into relief at the component level remains to be seen, but the cracks in the "build at all costs" strategy are becoming harder to ignore.